cephfs-data-scan: scrub tag filtering (#12133 and #12145) #5685
Conversation
|
This is the code that consumes the tags written by #5662 -- but it doesn't depend on that to merge. The two are tested together on the wip-scrub-jcs ceph-qa-suite branch, which can land once both parts are in master. |
|
@dachary looks like loic-bot hit a snag |
|
@jcsp it's a transient network error actually. Re-pushing will trigger the bot again. |
| // I want to select objects that have a name ending 00000000 | ||
| // and an xattr (scrub_tag) not equal to a specific value. | ||
| // This is so special case that we can't really pretend it's | ||
| // generic, so just fess up and call this the cephfs filter. |
There was a problem hiding this comment.
I think this is fine, but we could also have a generic "name and xattr" filter. Require it to match the name in a (definable, or just search-based) pattern, and to either match or not match an xattr name and value.
There was a problem hiding this comment.
When we're writing a plugin for the osd, I don't feel the need to make it customizable like this: it's already a special purpose call, and anyone who needs something more generic can write their own object class.
I'm going to take you at the "I think this is fine" part :-)
jcsp
force-pushed
the
wip-12133
branch
2 times, most recently
from
f607bed
to
9d7f441
Compare
|
Any objections to merging this @gregsfortytwo @ukernel ? Once this is in I can also merge ceph/ceph-qa-suite#709, which covers this along with the basic forward scrub functionality |
|
What testing have you subjected this to, @jcsp? It hasn't been through any integration branches — you just wrote the tests so I assume it's passing, but let's get that checked out first. |
|
This is tested in TestForwardScrub. You can see it passing here: Merge ceph/ceph-qa-suite#709 at the same time as this |
|
due to a osd bug (#6747), PGLSCephFSFilter code does not get executed. maybe we should merge this after the osd fix |
|
Updated for decode error handling as @ukernel suggests and retested in vstart vs. TestForwardScrub. Merged the other PR (good catch zheng!). In any case, this also works in its absence because it has a fallback path. |
No description provided.